Patent Inventorship Gets Clearer as AI Enters Drug Discovery
The U.S. Patent and Trademark Office issued revised guidance in late 2025 on who qualifies as an inventor when artificial intelligence plays a role in developing new drugs. The update clarifies that AI counts as a tool, not a co-inventor, which simplifies inventorship analysis for single-person teams but leaves questions open for collaborative research.
The new guidance treats AI systems like any other laboratory equipment. This means the person using the AI tool can claim inventorship without listing the AI itself as a contributor. That's a departure from earlier 2024 guidance, which required a more complex analysis borrowed from Federal Circuit precedent.
The Conception Standard Remains
Patent law defines inventorship through a single concept: conception. An inventor must form "a definite and permanent idea of the complete and operative invention" in their mind. For chemical compounds, this means knowing both the structure and how to make it.
The challenge lies in applying this definition to AI-assisted work. Drug discovery using AI often requires collaboration between computational experts and scientists. Determining who conceived of the final compound becomes complicated when multiple people contributed at different stages.
Predictive AI and Generative AI Are Different
The type of AI model used matters significantly for inventorship purposes.
Predictive AI takes existing data and forecasts outcomes-predicting how a known compound will behave, identifying drug targets, or revealing structure-activity relationships. When a scientist uses predictive AI to analyze a compound they've already identified and synthesized, the AI expert supervising that model likely won't qualify as a co-inventor. The AI expert must show they contributed significantly to the actual conception of the compound, not merely provided analysis of something already conceived.
Generative AI works differently. These models design entirely new molecular structures from scratch based on desired properties. A scientist specifies chemical features, and the model generates candidate compounds. Here, the person overseeing the generative AI model may contribute to conception of those newly generated compounds-especially if they understand the model's parameters and processes well enough to guide or troubleshoot the design process.
However, conception remains incomplete without a practical method to make the compound. If a scientist can only obtain a generative AI-designed compound with help from a chemist, all three people-the scientist, the AI expert, and the chemist-could qualify as co-inventors.
Joint Inventorship Still Requires Significant Contribution
When multiple people are involved, the Federal Circuit's test from Pannu v. Iolab Corp. still applies. To be a joint inventor, a person must:
- Contribute significantly to the conception or reduction to practice of the invention
- Contribute in a manner that isn't insignificant in quality relative to the full invention
- Do more than explain well-known concepts or provide standard information
Simply suggesting a desired result doesn't count as inventorship. Neither does providing textbook knowledge or explaining the state of the art. The contribution must be to the actual means of accomplishing the invention, not just the goal.
Documentation Is Critical
Companies using AI in drug discovery should maintain detailed records showing how each named inventor contributed to conception. Scientists typically document lab work and data analysis but rarely track the mental processes, computer workflows, and ideas fed into AI systems.
This documentation gap creates real risk. Patent disputes often turn on whether a filing contains a defensible record of human conception or relies on undocumented AI output. That distinction can determine whether a patent is enforceable or worthless as a commercial asset.
Records should document which scientist provided what information to the AI system and how that led to the final compound. When multiple inventors are named, the records should show how each one satisfied the Pannu factors.
Companies should also assess whether an invention came from a predictive or generative model, since the type affects both conception and co-inventorship analysis. These considerations fit naturally into an overall intellectual property governance strategy.
As AI becomes more embedded in drug discovery, the difference between robust documentation and vague records will determine which patents survive challenge and which ones don't.
Your membership also unlocks:
